CN114580520A

CN114580520A - Monitoring system and monitoring method for secondary equipment of power system

Info

Publication number: CN114580520A
Application number: CN202210187795.6A
Authority: CN
Inventors: 黄小杰
Original assignee: Shanghai Weiqiucheng Electronics Co ltd
Current assignee: Shanghai Weiqiucheng Electronics Co ltd
Priority date: 2022-02-28
Filing date: 2022-02-28
Publication date: 2022-06-03

Abstract

The utility model relates to the field of secondary equipment monitoring, and particularly discloses a monitoring system and a monitoring method for secondary equipment of a power system, which respectively extract global high-dimensional associated features of various measured value data acquired by secondary equipment deployed on a single power grid node and various data of a plurality of power grid nodes through a context-based encoder model, and calculate the distance between the feature vectors to be used as a parameter topological feature matrix, further extract the associated information of a data sample existing due to feature information and irregular topological structure information by using a graph neural network, and calculate the tag class evolution factors between the measured value topological feature vectors and the measured value feature vectors to be used as a loss function value training model so as to improve the accuracy of classification. Therefore, each power grid node data can be compressed, and therefore monitoring accuracy and timeliness of secondary equipment are guaranteed.

Description

Monitoring system and monitoring method for secondary equipment of power system

Technical Field

The present application relates to the field of secondary device monitoring, and more particularly, to a monitoring system of a secondary device of an electric power system and a monitoring method thereof.

Background

The secondary equipment is equipment for monitoring, controlling and protecting the primary equipment in the power system and is not directly connected with the generation of electric energy. The secondary equipment comprises a voltmeter, an ammeter, a power meter and an electric energy meter, and is mainly used for obtaining relevant parameters of the power equipment and judging whether the power grid system runs stably.

However, as the number of the secondary equipment increases, the related monitoring data soars, the monitoring accuracy of the secondary equipment is reduced, and the monitoring time of the equipment is prolonged. Therefore, in order to compress the monitored data to ensure the monitoring accuracy and timeliness of the secondary equipment, a monitoring system of the secondary equipment of the power system is desired.

Disclosure of Invention

The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a monitoring system and a monitoring method of secondary equipment of an electric power system, global high-dimensional correlation characteristic extraction is respectively carried out on various measured value data acquired by secondary equipment deployed on a single power grid node and various data of a plurality of power grid nodes through a context-based encoder model, the distance between feature vectors is obtained through calculation and is used as a parameter topological characteristic matrix, further, a graph neural network is used for extracting correlation information of a data sample due to feature information and irregular topological structure information, and a label class evolutionary factor between the measured value topological characteristic vector and the measured value characteristic vector is calculated and used as a loss function value training model, so that the accuracy of classification is improved. Therefore, each power grid node data can be compressed, and the monitoring accuracy and timeliness of secondary equipment are guaranteed.

According to an aspect of the present application, there is provided a monitoring system of a secondary device of a power system, including:

a training module comprising:

the training data unit is used for acquiring various measured value data acquired by secondary equipment deployed at each power grid node in the power system, and the measured value data comprises voltage data, current data, temperature monitoring data and power flow monitoring data;

the first global coding unit is used for enabling various measurement value data of each power grid node to pass through a context-based coder model comprising an embedded layer to obtain a plurality of characteristic vectors, and cascading the plurality of characteristic vectors to obtain a measurement value characteristic vector of each power grid node;

the training data extraction unit is used for acquiring all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all power grid nodes in the power system;

the second global coding unit is used for enabling all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all power grid nodes to pass through the context-based coder model comprising the embedded layer respectively to obtain a plurality of voltage characteristic vectors, a plurality of current characteristic vectors, a plurality of temperature characteristic vectors and a plurality of power flow characteristic vectors;

the distance characteristic matrix generating unit is used for respectively calculating the distance between every two characteristic vectors in the plurality of voltage characteristic vectors, the plurality of current characteristic vectors, the plurality of temperature characteristic vectors and the plurality of power flow characteristic vectors so as to obtain a voltage distance characteristic matrix, a current distance characteristic matrix, a temperature distance characteristic matrix and a power flow distance characteristic matrix;

the parameter characteristic fusion unit is used for calculating a position-weighted sum of the voltage distance characteristic matrix, the current distance characteristic matrix, the temperature distance characteristic matrix and the power flow distance characteristic matrix to obtain a parameter topological characteristic matrix;

the two-dimensional splicing unit is used for splicing the measured value eigenvectors of each power grid node into a measured value characteristic matrix in a two-dimensional mode;

a graph neural network unit, configured to pass the measurement value matrix and the parameter topology matrix through a graph neural network, where the graph neural network generates a measurement value topology feature matrix including measurement value feature information and parameter topology feature information through learnable neural network parameters, and each row vector in the measurement value topology feature matrix corresponds to a measurement value topology feature vector of one grid node;

a single-node loss function value calculation unit, configured to calculate, as a loss function value corresponding to each of the power grid nodes, a tag class evolution factor between the measured value topological feature vector of each of the power grid nodes and the measured value feature vector of the power grid node, where the tag class evolution factor is a difference between a cross entropy value between the measured value feature vector and the measured value topological feature vector and a logarithmic function value obtained by dividing a compressed class probability value of the measured value feature vector by a class probability value of the measured value topological feature vector, and the compressed class probability values of the measured value feature vector and the measured value topological feature vector are obtained by a classifier; and

a global training unit for training the encoder model and the graph neural network with a weighted sum of the loss function values for each of the grid nodes as a global loss function value; and

an inference module comprising:

the monitoring data unit is used for acquiring various measured value data acquired by secondary equipment deployed at each power grid node in the power system, and the measured value data comprises voltage data, current data, temperature monitoring data and tide monitoring data;

a first feature encoding unit, configured to pass measurement data of each grid node through the context-based encoder model including the embedded layer, which is trained by a training module, to obtain a plurality of feature vectors, and concatenate the plurality of feature vectors to obtain a measurement feature vector of each grid node;

the monitoring data extraction unit is used for acquiring all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all power grid nodes in the power system;

a second feature coding unit, configured to pass all voltage data, all current data, all temperature monitoring data, and all power flow monitoring data of all power grid nodes through the context-based encoder model including the embedded layer, which is trained by the training module, to obtain a plurality of voltage feature vectors, a plurality of current feature vectors, a plurality of temperature feature vectors, and a plurality of power flow feature vectors;

the parameter topological characteristic calculating unit is used for respectively calculating the distance between every two eigenvectors in the plurality of voltage eigenvectors, the plurality of current eigenvectors, the plurality of temperature eigenvectors and the plurality of load flow eigenvectors to obtain a voltage distance eigenvector matrix, a current distance eigenvector matrix, a temperature distance eigenvector matrix and a load flow distance eigenvector matrix;

the parameter topology fusion unit is used for calculating a position-weighted sum of the voltage distance characteristic matrix, the current distance characteristic matrix, the temperature distance characteristic matrix and the power flow distance characteristic matrix to obtain a parameter topology characteristic matrix;

the matrix splicing unit is used for splicing the measured value characteristic vectors of each power grid node into a measured value characteristic matrix in a two-dimensional mode;

the neural network unit is used for enabling the measured value matrix and the parameter topological matrix to pass through a graph neural network trained by a training module, and the graph neural network generates a measured value topological characteristic matrix containing measured value characteristic information and parameter topological characteristic information through learnable neural network parameters, wherein each row vector in the measured value topological characteristic matrix corresponds to a measured value topological characteristic vector of one power grid node;

and the compression ratio generation unit is used for enabling the measured value topological characteristic vector of each power grid node to pass through a classifier respectively to obtain a classification result, and the classification result is used for representing the compression ratio of each power grid node.

Compared with the prior art, the monitoring system and the monitoring method of the secondary equipment of the power system respectively perform global high-dimensional associated feature extraction on various measured value data acquired by the secondary equipment deployed on a single power grid node and various data of a plurality of power grid nodes through a context-based encoder model, calculate the distance between the obtained feature vectors to be used as a parameter topological feature matrix, further extract associated information of a data sample due to feature information and irregular topological structure information by using a graph neural network, and calculate a label class evolutionary factor between the measured value topological feature vector and the measured value feature vector to be used as a loss function value training model so as to improve the accuracy of classification. Therefore, each power grid node data can be compressed, and therefore monitoring accuracy and timeliness of secondary equipment are guaranteed.

Drawings

The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.

Fig. 1 is a schematic view of a monitoring system of a secondary device of an electric power system according to an embodiment of the present application.

Fig. 2 is a block diagram of a monitoring system of a power system secondary device according to an embodiment of the present application.

Fig. 3A is a flowchart of a training phase in a monitoring method of a monitoring system of a secondary device of an electric power system according to an embodiment of the present application.

Fig. 3B is a flowchart of an inference phase in a monitoring method of a monitoring system of a secondary device of an electric power system according to an embodiment of the present application.

Fig. 4 is a schematic diagram of an architecture of a training phase in a monitoring method of a monitoring system of secondary equipment of an electric power system according to an embodiment of the present application.

Fig. 5 is a schematic configuration diagram of an inference stage in a monitoring method of a monitoring system of a secondary device of an electric power system according to an embodiment of the present application.

Detailed Description

Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.

Overview of a scene

As mentioned above, the secondary device is a device for monitoring, controlling and protecting the primary device in the power system, and is not directly connected to the generation of electric energy. The secondary equipment comprises a voltmeter, an ammeter, a power meter and an electric energy meter, and is mainly used for obtaining relevant parameters of the power equipment and judging whether the power grid system runs stably.

However, as the number of the secondary equipment increases, the related monitoring data is surging, the monitoring accuracy of the secondary equipment is reduced, and the monitoring time of the equipment is prolonged. Therefore, in order to compress the monitored data to ensure the monitoring accuracy and timeliness of the secondary equipment, a monitoring system of the secondary equipment of the power system is desired.

Therefore, in the technical scheme of the application, various measurement value data acquired by secondary equipment deployed in a single power grid node are firstly acquired, and the acquired feature vectors are cascaded through a context-based encoder model including an embedded layer to acquire the measurement value feature vectors. And obtaining voltage data of a plurality of power grid nodes, and calculating the distance between the obtained characteristic vectors through a context-based encoder model comprising an embedded layer to obtain a voltage distance characteristic matrix. Similarly, distance feature matrices of other data are obtained, and the weighted sum is calculated as a parameter topology feature matrix.

And (3) passing the measured value feature matrix and the parameter topological feature matrix which are formed by two-dimensionally splicing the measured value feature vectors through a graph neural network to obtain a measured value topological feature matrix, wherein each line of the measured value topological feature matrix is the measured value topological feature vector of a certain power grid node.

Since the compression ratio can be regarded as the compression class label probability, the measured value topological feature vector v is calculated₂And measured value feature vector v₁The label class evolutionary factor in between as a loss function value training model, the loss function value is expressed as:

P(v₁) And P (v)₂) Respectively represent v₁And v₂Compression class probability value of, i.e. v₁And v₂A class probability value obtained by the classifier.

In each iteration cycle, using the above v for each grid node₁And v₂The vectors calculate loss function values and calculate a weighted sum of the loss function values for each grid node as a global loss function value to update the model parameters.

When deducing, the topological feature vector v of the measured value is₂By the classifier, its class probability value is obtained as a compression ratio.

Based on this, the application provides a monitoring system of power system secondary equipment, and the monitoring system comprises a training module and an inference module. Wherein, the training module includes: the training data unit is used for acquiring various measured value data acquired by secondary equipment deployed at each power grid node in the power system, and the measured value data comprises voltage data, current data, temperature monitoring data and power flow monitoring data; the first global coding unit is used for enabling various measurement value data of each power grid node to pass through a context-based coder model comprising an embedded layer to obtain a plurality of characteristic vectors, and cascading the plurality of characteristic vectors to obtain a measurement value characteristic vector of each power grid node; the training data extraction unit is used for acquiring all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all power grid nodes in the power system; the second global coding unit is used for enabling all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all power grid nodes to pass through the context-based coder model comprising the embedded layer respectively to obtain a plurality of voltage characteristic vectors, a plurality of current characteristic vectors, a plurality of temperature characteristic vectors and a plurality of power flow characteristic vectors; the distance characteristic matrix generating unit is used for respectively calculating the distance between every two characteristic vectors in the plurality of voltage characteristic vectors, the plurality of current characteristic vectors, the plurality of temperature characteristic vectors and the plurality of power flow characteristic vectors so as to obtain a voltage distance characteristic matrix, a current distance characteristic matrix, a temperature distance characteristic matrix and a power flow distance characteristic matrix; the parameter characteristic fusion unit is used for calculating a position-weighted sum of the voltage distance characteristic matrix, the current distance characteristic matrix, the temperature distance characteristic matrix and the power flow distance characteristic matrix to obtain a parameter topological characteristic matrix; the two-dimensional splicing unit is used for splicing the measured value eigenvectors of each power grid node into a measured value characteristic matrix in a two-dimensional mode; a graph neural network unit, configured to pass the measurement value matrix and the parameter topology matrix through a graph neural network, where the graph neural network generates a measurement value topology feature matrix including measurement value feature information and parameter topology feature information through learnable neural network parameters, and each row vector in the measurement value topology feature matrix corresponds to a measurement value topology feature vector of one grid node; a single-node loss function value calculation unit, configured to calculate, as a loss function value corresponding to each of the power grid nodes, a tag class evolution factor between the measured value topological feature vector of each of the power grid nodes and the measured value feature vector of the power grid node, where the tag class evolution factor is a difference between a cross entropy value between the measured value feature vector and the measured value topological feature vector and a logarithmic function value obtained by dividing a compressed class probability value of the measured value feature vector by a class probability value of the measured value topological feature vector, and the compressed class probability values of the measured value feature vector and the measured value topological feature vector are obtained by a classifier; and a global training unit for training the encoder model and the graph neural network with a weighted sum of the loss function values of each of the grid nodes as a global loss function value. Wherein, the inference module comprises: the monitoring data unit is used for acquiring various measured value data acquired by secondary equipment deployed at each power grid node in the power system, and the measured value data comprises voltage data, current data, temperature monitoring data and tide monitoring data; a first feature encoding unit, configured to pass measurement data of each grid node through the context-based encoder model including the embedded layer, which is trained by a training module, to obtain a plurality of feature vectors, and cascade the plurality of feature vectors to obtain a measurement feature vector of each grid node; the monitoring data extraction unit is used for acquiring all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all power grid nodes in the power system; a second feature coding unit, configured to pass all voltage data, all current data, all temperature monitoring data, and all power flow monitoring data of all power grid nodes through the context-based encoder model including the embedded layer, which is trained by the training module, to obtain a plurality of voltage feature vectors, a plurality of current feature vectors, a plurality of temperature feature vectors, and a plurality of power flow feature vectors; the parameter topological characteristic calculating unit is used for respectively calculating the distance between every two eigenvectors in the plurality of voltage eigenvectors, the plurality of current eigenvectors, the plurality of temperature eigenvectors and the plurality of load flow eigenvectors to obtain a voltage distance eigenvector matrix, a current distance eigenvector matrix, a temperature distance eigenvector matrix and a load flow distance eigenvector matrix; the parameter topology fusion unit is used for calculating a position-weighted sum of the voltage distance characteristic matrix, the current distance characteristic matrix, the temperature distance characteristic matrix and the power flow distance characteristic matrix to obtain a parameter topology characteristic matrix; the matrix splicing unit is used for splicing the measured value characteristic vectors of each power grid node into a measured value characteristic matrix in a two-dimensional mode; the neural network unit is used for enabling the measured value matrix and the parameter topological matrix to pass through a graph neural network trained by a training module, and the graph neural network generates a measured value topological characteristic matrix containing measured value characteristic information and parameter topological characteristic information through learnable neural network parameters, wherein each row vector in the measured value topological characteristic matrix corresponds to a measured value topological characteristic vector of one power grid node; and the compression ratio generation unit is used for enabling the measured value topological characteristic vector of each power grid node to pass through a classifier respectively to obtain a classification result, and the classification result is used for representing the compression ratio of each power grid node.

Fig. 1 illustrates a scene schematic diagram of a monitoring system of a secondary device of a power system according to an embodiment of the present application. As shown in fig. 1, in the training phase of the application scenario, first, measurement data, including voltage data, current data, temperature monitoring data, and power flow monitoring data, are collected by secondary devices (e.g., T as illustrated in fig. 1) deployed at each grid node (e.g., G as illustrated in fig. 1) in the power system (e.g., P as illustrated in fig. 1), including but not limited to a voltage meter, a current meter, a power meter, and all voltage data, all current data, all temperature monitoring data, and all power flow monitoring data of all grid nodes in the power system are obtained. Then, the obtained measurement data items, and all voltage data, all current data, all temperature monitoring data, and all power flow monitoring data of all grid nodes in the power system are input into a server (for example, S as illustrated in fig. 1) in which a monitoring algorithm of a power system secondary device is deployed, wherein the server is capable of training the encoder model and the graph neural network of the monitoring system of the power system secondary device with the measurement data items, and all voltage data, all current data, all temperature monitoring data, and all power flow monitoring data of all grid nodes in the power system based on the monitoring algorithm of the power system secondary device.

After the training is completed, in the inference phase, first, measurement data items are collected by secondary devices (for example, T as illustrated in fig. 1) deployed at each grid node (for example, G as illustrated in fig. 1) in the power system (for example, P as illustrated in fig. 1), and all voltage data, all current data, all temperature monitoring data, and all power flow monitoring data of all grid nodes in the power system are acquired. Then, the measurement value data, and all voltage data, all current data, all temperature monitoring data, and all power flow monitoring data of all grid nodes in the power system are input into a server (for example, S as illustrated in fig. 1) in which a monitoring algorithm of a power system secondary device is deployed, wherein the server is capable of processing the measurement value data, and all voltage data, all current data, all temperature monitoring data, and all power flow monitoring data of all grid nodes in the power system with the monitoring algorithm of the power system secondary device to generate a classification result representing a compression ratio of each grid node.

Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.

Exemplary System

Fig. 2 illustrates a block diagram of a monitoring system of a power system secondary device according to an embodiment of the present application. As shown in fig. 2, a monitoring system 200 of a secondary device of a power system according to an embodiment of the present application includes: a training module 210 and an inference module 220. Wherein, the training module 210 includes: a training data unit 2101 configured to obtain various measurement value data acquired by secondary devices deployed at each grid node in the power system, where the measurement value data includes voltage data, current data, temperature monitoring data, and power flow monitoring data; a first global encoding unit 2102, configured to pass measurement value data of each of the grid nodes through a context-based encoder model including an embedded layer to obtain a plurality of feature vectors, and concatenate the plurality of feature vectors to obtain a measurement value feature vector of each of the grid nodes; a training data extraction unit 2103, configured to obtain all voltage data, all current data, all temperature monitoring data, and all power flow monitoring data of all power grid nodes in the power system; a second global encoding unit 2104 for passing all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all grid nodes through the context-based encoder model including the embedding layer to obtain a plurality of voltage feature vectors, a plurality of current feature vectors, a plurality of temperature feature vectors and a plurality of power flow feature vectors, respectively; a distance feature matrix generating unit 2105, configured to calculate distances between every two feature vectors in the plurality of voltage feature vectors, the plurality of current feature vectors, the plurality of temperature feature vectors, and the plurality of power flow feature vectors, respectively, to obtain a voltage distance feature matrix, a current distance feature matrix, a temperature distance feature matrix, and a power flow distance feature matrix; a parameter feature fusion unit 2106, configured to calculate a position-weighted sum between the voltage distance feature matrix, the current distance feature matrix, the temperature distance feature matrix, and the power flow distance feature matrix to obtain a parameter topology feature matrix; a two-dimensional splicing unit 2107, configured to splice the measured value eigenvectors of each power grid node into a measured value eigenvector matrix in a two-dimensional manner; a graph neural network unit 2108, configured to pass the measurement value matrix and the parameter topology matrix through a graph neural network, which generates a measurement value topology feature matrix including measurement value feature information and parameter topology feature information through learnable neural network parameters, wherein each row vector in the measurement value topology feature matrix corresponds to a measurement value topology feature vector of one grid node; a single-node loss function value calculation unit 2109, configured to calculate, as a loss function value corresponding to each of the power grid nodes, a tag class evolution factor between the measured value topological feature vector of each of the power grid nodes and the measured value feature vector of the power grid node, where the tag class evolution factor is a difference between a cross entropy value between the measured value feature vector and the measured value topological feature vector and a logarithmic function value obtained by dividing a compressed class probability value of the measured value feature vector by a class probability value of the measured value topological feature vector, where the compressed class probability values of the measured value feature vector and the measured value topological feature vector are obtained by the classifier; and a global training unit 2110 for training the encoder model and the graph neural network with a weighted sum of the loss function values of each of the grid nodes as a global loss function value. The inference module 220 includes: the monitoring data unit 221 is configured to acquire measurement value data of each item acquired by secondary equipment deployed in each grid node in the power system, where the measurement value data includes voltage data, current data, temperature monitoring data, and power flow monitoring data; a first feature encoding unit 222, configured to pass measurement data of each of the grid nodes through the context-based encoder model including the embedded layer, which is trained by a training module, to obtain a plurality of feature vectors, and concatenate the plurality of feature vectors to obtain a measurement feature vector of each of the grid nodes; the monitoring data extraction unit 223 is configured to obtain all voltage data, all current data, all temperature monitoring data, and all power flow monitoring data of all power grid nodes in the power system; a second feature encoding unit 224, configured to pass all voltage data, all current data, all temperature monitoring data, and all power flow monitoring data of all grid nodes through the context-based encoder model including an embedded layer, which is trained by the training module, to obtain a plurality of voltage feature vectors, a plurality of current feature vectors, a plurality of temperature feature vectors, and a plurality of power flow feature vectors; a parameter topology feature calculation unit 225, configured to calculate distances between every two feature vectors in the plurality of voltage feature vectors, the plurality of current feature vectors, the plurality of temperature feature vectors, and the plurality of power flow feature vectors, respectively, to obtain a voltage distance feature matrix, a current distance feature matrix, a temperature distance feature matrix, and a power flow distance feature matrix; a parameter topology fusion unit 226, configured to calculate a position-weighted sum between the voltage distance feature matrix, the current distance feature matrix, the temperature distance feature matrix, and the power flow distance feature matrix to obtain a parameter topology feature matrix; the matrix splicing unit 227 is used for splicing the measured value eigenvectors of each power grid node into a measured value eigenvector matrix in a two-dimensional manner; a neural network unit 228, configured to pass the measurement value matrix and the parameter topology matrix through a graph neural network trained by a training module, where the graph neural network generates a measurement value topology feature matrix including measurement value feature information and parameter topology feature information through learnable neural network parameters, where each row vector in the measurement value topology feature matrix corresponds to a measurement value topology feature vector of one grid node; the compression ratio generation unit 229 is configured to pass the topology feature vector of the measurement value of each of the grid nodes through a classifier to obtain a classification result, where the classification result is used to represent a compression ratio of each of the grid nodes.

Specifically, in the embodiment of the present application, in the training module 210, the training data unit 2101 and the first global coding unit 2102 are configured to obtain measurement value data collected by secondary devices deployed on each grid node in the power system, where the measurement value data includes voltage data, current data, temperature monitoring data, and power flow monitoring data, pass the measurement value data of each grid node through a context-based encoder model including an embedded layer to obtain a plurality of feature vectors, and cascade the feature vectors to obtain a measurement value feature vector of each grid node. As described above, as the number of the secondary devices increases, the related monitoring data soars, the monitoring accuracy of the secondary devices is reduced, and the monitoring time of the devices is also prolonged, in the technical solution of the present application, it is expected that data compression is performed comprehensively by using the voltage data, the current data, the temperature monitoring data, and the power flow monitoring data collected by the secondary devices deployed at each grid node in the power system, so as to improve the monitoring accuracy and timeliness of the secondary devices.

That is, specifically, in the technical solution of the present application, first, various pieces of measured value data are collected by secondary devices deployed at various grid nodes in the power system, where the measured value data include voltage data, current data, temperature monitoring data, and power flow monitoring data, and the secondary devices include, but are not limited to, a voltmeter, an ammeter, a power meter, and an electric energy meter. Then, each item of measurement value data of each power grid node is encoded in a context-based encoder model comprising an embedded layer, so that global correlation information of each item of measurement value data is extracted, and a plurality of feature vectors are obtained. And further cascading the plurality of characteristic vectors to integrate high-dimensional associated characteristic information of each measured value data of each power grid node, so as to obtain the measured value characteristic vector of each power grid node.

More specifically, in an embodiment of the present application, the first global encoding unit includes: the embedding subunit is used for enabling each item of measured value data of each power grid node to respectively pass through an embedding layer of the encoder model so as to convert each item of measured value into an input vector to obtain a sequence of input vectors; a converter global encoding subunit for passing the sequence of input vectors through a converter of the encoder model to obtain the plurality of feature vectors; and the cascade subunit is used for cascading the plurality of characteristic vectors to obtain a measured value characteristic vector of each power grid node.

Specifically, in the embodiment of the present application, in the training module 210, the training data extraction unit 2103 and the second global encoding unit 2104 are configured to obtain all voltage data, all current data, all temperature monitoring data, and all power flow monitoring data of all grid nodes in the power system, and pass all voltage data, all current data, all temperature monitoring data, and all power flow monitoring data of all grid nodes through the context-based encoder model including the embedded layer to obtain a plurality of voltage feature vectors, a plurality of current feature vectors, a plurality of temperature feature vectors, and a plurality of power flow feature vectors. That is, in the solution of the present application, similarly, all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of a plurality of grid nodes in the power system are processed through a context-based encoder model including an embedded layer to obtain a plurality of voltage eigenvectors, a plurality of current eigenvectors, a plurality of temperature eigenvectors and a plurality of power flow eigenvectors with global information.

Specifically, in the embodiment of the present application, in the training module 210, the distance feature matrix generating unit 2105 and the parameter feature fusing unit 2106 are configured to calculate distances between every two feature vectors of the plurality of voltage feature vectors, the plurality of current feature vectors, the plurality of temperature feature vectors, and the plurality of power flow feature vectors to obtain a voltage distance feature matrix, a current distance feature matrix, a temperature distance feature matrix, and a power flow distance feature matrix, and calculate a position-weighted sum between the voltage distance feature matrix, the current distance feature matrix, the temperature distance feature matrix, and the power flow distance feature matrix to obtain a parameter topological feature matrix, respectively. That is, in the technical solution of the present application, distances between every two eigenvectors in the plurality of voltage eigenvectors, the plurality of current eigenvectors, the plurality of temperature eigenvectors, and the plurality of power flow eigenvectors are further respectively calculated to obtain a voltage distance eigenvector matrix, a current distance eigenvector matrix, a temperature distance eigenvector matrix, and a power flow distance eigenvector matrix, so as to respectively express similarity relations among the voltage characteristics, the current characteristics, the temperature characteristics, and the power flow characteristics. Then, the weighted sum is calculated to be used as a parameter topological characteristic matrix so as to integrate characteristic information and facilitate subsequent implicit characteristic extraction.

More specifically, in this embodiment of the present application, the distance feature matrix generating unit includes: respectively calculating cosine distances between every two eigenvectors in the plurality of voltage eigenvectors, and arranging the Euclidean distances into the voltage distance eigenvector matrix; respectively calculating cosine distances between every two eigenvectors in the plurality of current eigenvectors, and arranging the Euclidean distances into the current distance eigenvector matrix; respectively calculating cosine distances between every two eigenvectors in the plurality of temperature eigenvectors, and arranging the Euclidean distances into the temperature distance eigenvector matrix; and respectively calculating cosine distances between every two characteristic vectors in the plurality of power flow characteristic vectors, and arranging the Euclidean distances into the power flow distance characteristic matrix.

Specifically, in the embodiment of the present application, in the training module 210, the two-dimensional stitching unit 2107 and the graph neural network unit 2108 are configured to two-dimensionally stitch the measurement value feature vectors of each of the grid nodes into a measurement value feature matrix, and pass the measurement value matrix and the parameter topology matrix through a graph neural network, which generates a measurement value topology feature matrix including measurement value feature information and parameter topology feature information through learnable neural network parameters, where each row vector in the measurement value topology feature matrix corresponds to a measurement value topology feature vector of one grid node. Namely, the obtained measured value eigenvectors of each power grid node are further spliced into a measured value eigenvector matrix in two dimensions. Then, the measured value matrix and the parameter topological matrix are processed through a graph neural network, wherein the graph neural network generates a measured value topological characteristic matrix containing measured value characteristic information and parameter topological characteristic information through learnable neural network parameters. It should be understood that the graph neural network can be used for processing graph data in an irregular non-euclidean space, so that associated information of a data sample due to measured value feature information and irregular parameter topological structure information can be extracted, and therefore, compared with a feature matrix obtained by directly splicing obtained measured value topological feature matrices, the obtained measured value topological feature matrices can improve the accuracy of classification.

Specifically, in the embodiment of the present application, in the training module 210, the single-node loss function value calculating unit 2109 and the global training unit 2110 are configured to calculate, as the loss function value corresponding to each of the grid nodes, a label category evolution factor between the measured value topological feature vector of each of the grid nodes and the measured value feature vector of the grid node with respect to the measured value feature vector of the grid node, where the label category evolution factor is a difference between a cross entropy value between the measured value feature vector and the measured value topological feature vector and a logarithmic function value obtained by dividing a compression category probability value of the measured value feature vector by a category probability value of the measured value topological feature vector, and the compression category probability values of the measured value topological feature vector and the measured value topological feature vector are obtained by a classifier, and training the encoder model and the graph neural network with a weighted sum of the loss function values for each of the grid nodes as a global loss function value. It should be understood that, because the compression ratio can be regarded as the compression class label probability, in the technical solution of the present application, the measurement topological feature vector v is calculated₂And the measured value feature vector v₁And (4) the label class evolution factors in between are used as loss function value training models. And, in each iteration cycle, using said v for each said grid node₁And v₂Vector-computing loss function values and computing a weighted sum of the loss function values for each of the grid nodes as a global loss function value to update the model parameters.

More specifically, in an embodiment of the present application, the single-node loss function value calculating unit is further configured to: calculating a label category evolution factor between the measured value topological characteristic vector of each power grid node relative to the measured value characteristic vector of the power grid node as a loss function value corresponding to each power grid node according to the following formula; the formula is:

wherein v is₂For topological feature vectors of measured values, v₁For the measured value feature vector, P (v)₁) And P (v)₂) Respectively represent v₁And v₂Compression class probability value of, i.e. v₁And v₂A class probability value obtained by the classifier.

In particular, in a specific example, the single-node loss function value calculating unit is further configured to: respectively passing the measured value topological feature vector and the measured value feature vector through a classifier, wherein the classifier respectively processes the measured value topological feature vector and the measured value feature vector to generate a compressed class probability value of the measured value topological feature vector and a compressed class probability value of the measured value feature vector according to the following formulas: softmax { (W)_n,B_n):…:(W₁,B₁)|X}。

After the training is completed, the measured value topological characteristic matrix can be obtained according to the method in the inference module. And then, respectively passing the measured value topological characteristic vector of each power grid node through a classifier to obtain a compression ratio classification result for representing each power grid node.

In summary, the monitoring system 200 for the secondary device of the power system according to the embodiment of the present application is illustrated, which performs global high-dimensional associated feature extraction on each item of measured value data acquired by the secondary device deployed in a single power grid node and each item of data of multiple power grid nodes through a context-based encoder model, calculates a distance between the obtained feature vectors to serve as a parameter topological feature matrix, further extracts associated information of a data sample due to feature information and irregular topological structure information by using a graph neural network, and calculates a label class evolutionary factor between the measured value topological feature vector and the measured value feature vector to serve as a loss function value training model, so as to improve the accuracy of classification. Therefore, each power grid node data can be compressed, and the monitoring accuracy and timeliness of secondary equipment are guaranteed.

As described above, the monitoring system 200 of the power system secondary device according to the embodiment of the present application can be implemented in various terminal devices, such as a server of a monitoring algorithm of the power system secondary device, and the like. In one example, the monitoring system 200 of the power system secondary device according to the embodiment of the present application may be integrated into the terminal device as one software module and/or hardware module. For example, the monitoring system 200 of the power system secondary device may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the monitoring system 200 of the secondary device of the power system may also be one of many hardware modules of the terminal device.

Alternatively, in another example, the monitoring system 200 of the power system secondary device and the terminal device may also be separate devices, and the monitoring system 200 of the power system secondary device may be connected to the terminal device through a wired and/or wireless network and transmit the mutual information according to an agreed data format.

Exemplary method

Fig. 3A illustrates a flowchart of a training phase in a monitoring method of a monitoring system of a secondary device of a power system according to an embodiment of the present application. As shown in fig. 3A, a monitoring method of a monitoring system of a secondary device of an electric power system according to an embodiment of the present application includes: a training phase comprising the steps of: s110, acquiring various measured value data collected by secondary equipment deployed at each power grid node in the power system, wherein the measured value data comprises voltage data, current data, temperature monitoring data and power flow monitoring data; s120, enabling each item of measured value data of each power grid node to pass through a context-based encoder model comprising an embedded layer to obtain a plurality of characteristic vectors, and cascading the plurality of characteristic vectors to obtain a measured value characteristic vector of each power grid node; s130, acquiring all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all power grid nodes in the power system; s140, respectively passing all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all power grid nodes through the context-based encoder model comprising the embedded layer to obtain a plurality of voltage characteristic vectors, a plurality of current characteristic vectors, a plurality of temperature characteristic vectors and a plurality of power flow characteristic vectors; s150, respectively calculating the distance between every two eigenvectors in the plurality of voltage eigenvectors, the plurality of current eigenvectors, the plurality of temperature eigenvectors and the plurality of load flow eigenvectors to obtain a voltage distance eigenvector matrix, a current distance eigenvector matrix, a temperature distance eigenvector matrix and a load flow distance eigenvector matrix; s160, calculating a position-weighted sum of the voltage distance characteristic matrix, the current distance characteristic matrix, the temperature distance characteristic matrix and the power flow distance characteristic matrix to obtain a parameter topological characteristic matrix; s170, two-dimensionally splicing the measured value eigenvectors of each power grid node into a measured value eigenvector matrix; s180, passing the measured value matrix and the parameter topological matrix through a graph neural network, wherein the graph neural network generates a measured value topological characteristic matrix containing measured value characteristic information and parameter topological characteristic information through learnable neural network parameters, and each row vector in the measured value topological characteristic matrix corresponds to a measured value topological characteristic vector of one power grid node; s190, calculating a label category evolution factor between the measured value topological feature vector of each power grid node and the measured value feature vector of each power grid node relative to each power grid node as a loss function value corresponding to each power grid node, wherein the label category evolution factor is a difference value between a cross entropy value between the measured value feature vector and the measured value topological feature vector and a logarithmic function value obtained by dividing a compressed category probability value of the measured value feature vector by the category probability value of the measured value topological feature vector, and the compressed category probability values of the measured value feature vector and the measured value topological feature vector are obtained by a classifier; and S200, training the encoder model and the graph neural network by taking the weighted sum of the loss function values of each power grid node as a global loss function value.

Fig. 3B illustrates a flow chart of an inference phase in a monitoring method of a monitoring system of a power system secondary device according to an embodiment of the present application. As shown in fig. 3B, a monitoring method of a monitoring system of a secondary device in an electric power system according to an embodiment of the present application includes: an inference phase comprising the steps of: s210, acquiring various measured value data collected by secondary equipment deployed in each power grid node in the power system, wherein the measured value data comprises voltage data, current data, temperature monitoring data and power flow monitoring data; s220, passing each measured value data of each power grid node through the context-based encoder model containing the embedded layer and trained by the training module to obtain a plurality of feature vectors, and cascading the plurality of feature vectors to obtain a measured value feature vector of each power grid node; s230, acquiring all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all power grid nodes in the power system; s240, respectively passing all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all power grid nodes through the context-based encoder model containing the embedded layer and trained by the training module to obtain a plurality of voltage feature vectors, a plurality of current feature vectors, a plurality of temperature feature vectors and a plurality of power flow feature vectors; s250, respectively calculating the distance between every two eigenvectors in the plurality of voltage eigenvectors, the plurality of current eigenvectors, the plurality of temperature eigenvectors and the plurality of load flow eigenvectors to obtain a voltage distance eigenvector matrix, a current distance eigenvector matrix, a temperature distance eigenvector matrix and a load flow distance eigenvector matrix; s260, calculating a position-weighted sum of the voltage distance characteristic matrix, the current distance characteristic matrix, the temperature distance characteristic matrix and the power flow distance characteristic matrix to obtain a parameter topological characteristic matrix; s270, splicing the measured value eigenvectors of each power grid node into a measured value eigenvector two-dimensionally; s280, the measured value matrix and the parameter topological matrix pass through a graph neural network trained by a training module, and the graph neural network generates a measured value topological characteristic matrix containing measured value characteristic information and parameter topological characteristic information through learnable neural network parameters, wherein each row vector in the measured value topological characteristic matrix corresponds to a measured value topological characteristic vector of one power grid node; s290, the measured value topological characteristic vector of each power grid node is respectively passed through a classifier to obtain a classification result, and the classification result is used for representing the compression ratio of each power grid node.

Fig. 4 illustrates an architecture diagram of a training phase in a monitoring method of a monitoring system of a secondary device of a power system according to an embodiment of the present application. As shown in fig. 4, in the training phase, first, the measurement value data (e.g., P1 as illustrated in fig. 4) of each grid node is passed through a context-based encoder model (e.g., E1 as illustrated in fig. 4) including an embedding layer to obtain a plurality of feature vectors (e.g., FV1 as illustrated in fig. 4), and the plurality of feature vectors are concatenated to obtain a measurement value feature vector (e.g., FV2 as illustrated in fig. 4) of each grid node; then, passing all obtained voltage data (e.g., Q1 as illustrated in fig. 4), all current data (e.g., Q2 as illustrated in fig. 4), all temperature monitoring data (e.g., Q3 as illustrated in fig. 4) and all power flow monitoring data (e.g., Q4 as illustrated in fig. 4) of the all grid nodes through the context-based encoder model including an embedding layer (e.g., E1 as illustrated in fig. 4) to obtain a plurality of voltage eigenvectors (e.g., VF1 as illustrated in fig. 4), a plurality of current eigenvectors (e.g., VF2 as illustrated in fig. 4), a plurality of temperature eigenvectors (e.g., VF3 as illustrated in fig. 4) and a plurality of power flow eigenvectors (e.g., VF4 as illustrated in fig. 4), respectively; then, calculating distances between each two eigenvectors of the plurality of voltage eigenvectors, the plurality of current eigenvectors, the plurality of temperature eigenvectors, and the plurality of power flow eigenvectors to obtain a voltage distance eigenvector matrix (e.g., MF1 as illustrated in fig. 4), a current distance eigenvector matrix (e.g., MF2 as illustrated in fig. 4), a temperature distance eigenvector matrix (e.g., MF3 as illustrated in fig. 4), and a power flow distance eigenvector matrix (e.g., MF4 as illustrated in fig. 4), respectively; then, calculating a position-weighted sum between the voltage distance feature matrix, the current distance feature matrix, the temperature distance feature matrix and the power flow distance feature matrix to obtain a parameter topology feature matrix (e.g., MF as illustrated in fig. 4); then, two-dimensionally splicing the measurement feature vectors of each of the grid nodes into a measurement feature matrix (e.g., FM as illustrated in fig. 4); next, passing the measurement value matrix and the parameter topology matrix through a graph neural network (e.g., GNN as illustrated in fig. 4) that generates a measurement value topology feature matrix (e.g., M as illustrated in fig. 4) including measurement value feature information and parameter topology feature information from learnable neural network parameters; then, calculating a label class evolution factor (for example, GEF as illustrated in fig. 4) between the measured value topological characteristic vector of each power grid node relative to the measured value characteristic vector of the power grid node as a loss function value corresponding to each power grid node; and finally, training the encoder model and the graph neural network with a weighted sum of the loss function values for each of the grid nodes as a global loss function value.

Fig. 5 illustrates an architecture diagram of an inference phase in a monitoring method of a monitoring system of a secondary device of a power system according to an embodiment of the present application. As shown in fig. 5, in the inference phase, first, the measurement value data (e.g., P1 as illustrated in fig. 5) of each grid node is passed through the context-based encoder model including an embedding layer (e.g., E2 as illustrated in fig. 5) trained by the training module to obtain a plurality of feature vectors (e.g., FV1 as illustrated in fig. 5), and the plurality of feature vectors are concatenated to obtain the measurement value feature vector (e.g., FV2 as illustrated in fig. 5) of each grid node; then, passing all the obtained voltage data (e.g., Q1 as illustrated in fig. 5), all the obtained current data (e.g., Q2 as illustrated in fig. 5), all the obtained temperature monitoring data (e.g., Q3 as illustrated in fig. 5) and all the obtained power flow monitoring data (e.g., Q4 as illustrated in fig. 5) through the context-based encoder model (e.g., E2 as illustrated in fig. 5) including the embedded layer trained by the training module to obtain a plurality of voltage eigenvectors (e.g., VF1 as illustrated in fig. 5), a plurality of current eigenvectors (e.g., VF2 as illustrated in fig. 5), a plurality of temperature eigenvectors (e.g., VF3 as illustrated in fig. 5) and a plurality of power flow eigenvectors (e.g., VF4 as illustrated in fig. 5), respectively; then, calculating distances between each two eigenvectors of the plurality of voltage eigenvectors, the plurality of current eigenvectors, the plurality of temperature eigenvectors, and the plurality of power flow eigenvectors to obtain a voltage distance eigenvector matrix (e.g., MF1 as illustrated in fig. 5), a current distance eigenvector matrix (e.g., MF2 as illustrated in fig. 5), a temperature distance eigenvector matrix (e.g., MF3 as illustrated in fig. 5), and a power flow distance eigenvector matrix (e.g., MF4 as illustrated in fig. 5), respectively; then, calculating a position-weighted sum between the voltage distance feature matrix, the current distance feature matrix, the temperature distance feature matrix and the power flow distance feature matrix to obtain a parameter topology feature matrix (e.g., MF as illustrated in fig. 5); then, two-dimensionally splicing the measurement feature vectors of each of the grid nodes into a measurement feature matrix (e.g., FM as illustrated in fig. 5); then, passing the measured value matrix and the parameter topology matrix through a graph neural network (e.g., GN as illustrated in FIG. 5) trained by a training module, the graph neural network generating a measured value topology feature matrix (e.g., M as illustrated in FIG. 5) including measured value feature information and parameter topology feature information through learnable neural network parameters; finally, the measured value topological feature vector of each grid node is respectively passed through a classifier (for example, a classifier as illustrated in fig. 5) to obtain a classification result, and the classification result is used for representing the compression ratio of each grid node.

In summary, the monitoring method of the monitoring system of the secondary device of the power system according to the embodiment of the present application is clarified, and global high-dimensional associated feature extraction is performed on each item of measured value data acquired by the secondary device deployed in a single power grid node and each item of data of multiple power grid nodes by using a context-based encoder model, and a distance between the feature vectors is calculated to be used as a parameter topological feature matrix, and further, associated information existing in a data sample due to feature information and irregular topological structure information is extracted by using a graph neural network, and a label class evolution factor between the measured value topological feature vector and the measured value feature vector is calculated to be used as a loss function value training model, so as to improve the accuracy of classification. Therefore, each power grid node data can be compressed, and the monitoring accuracy and timeliness of secondary equipment are guaranteed.

The foregoing describes the general principles of the present application in conjunction with specific embodiments, however, it is noted that the advantages, effects, etc. mentioned in the present application are merely examples and are not limiting, and they should not be considered essential to the various embodiments of the present application. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the foregoing disclosure is not intended to be exhaustive or to limit the disclosure to the precise details disclosed.

The block diagrams of devices, apparatuses, systems referred to in this application are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".

It should also be noted that in the devices, apparatuses, and methods of the present application, the components or steps may be decomposed and/or recombined. These decompositions and/or recombinations are to be considered as equivalents of the present application.

The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A monitoring system for a secondary device of an electrical power system, comprising:

a training module comprising:

an inference module comprising:

2. The monitoring system of the power system secondary device according to claim 1, wherein the first global encoding unit includes:

the embedding subunit is used for enabling each item of measured value data of each power grid node to respectively pass through an embedding layer of the encoder model so as to convert each item of measured value into an input vector to obtain a sequence of input vectors;

a converter global encoding subunit for passing the sequence of input vectors through a converter of the encoder model to obtain the plurality of feature vectors; and

and the cascade subunit is used for cascading the plurality of characteristic vectors to obtain a measured value characteristic vector of each power grid node.

3. The monitoring system of the power system secondary device according to claim 2, wherein the distance feature matrix generation unit is further configured to:

respectively calculating cosine distances between every two eigenvectors in the plurality of voltage eigenvectors, and arranging the Euclidean distances into the voltage distance eigenvector matrix;

respectively calculating cosine distances between every two eigenvectors in the plurality of current eigenvectors, and arranging the Euclidean distances into the current distance eigenvector matrix;

respectively calculating cosine distances between every two eigenvectors in the plurality of temperature eigenvectors, and arranging the Euclidean distances into the temperature distance eigenvector matrix; and

respectively calculating cosine distances between every two characteristic vectors in the plurality of power flow characteristic vectors, and arranging the Euclidean distances into the power flow distance characteristic matrix.

4. The monitoring system of the power system secondary device according to claim 3, wherein the single-node loss function value calculation unit is further configured to calculate a label category evolution factor between the measured value topological feature vector of each of the grid nodes with respect to the measured value feature vector of the grid node as the loss function value corresponding to each of the grid nodes in the following formula;

the formula is:

wherein v is₂For topological feature vectors of measured values, v₁For the measured value feature vector, P (v)₁) And P (v)₂) Respectively represent v₁And v₂The compression category probability value of.

5. According to the rightThe monitoring system of the power system secondary device according to claim 4, wherein the single-node loss function value calculation unit is further configured to pass the measured value topological feature vector and the measured value feature vector through a classifier, respectively, wherein the classifier processes the measured value topological feature vector and the measured value feature vector to generate a compressed class probability value of the measured value topological feature vector and a compressed class probability value of the measured value feature vector, respectively, according to the following formulas: softmax { (W)_n,B_n):…:(W₁,B₁)|X}。

6. A monitoring method of a monitoring system of a secondary device of a power system, characterized by comprising:

a training phase comprising:

acquiring various measurement value data acquired by secondary equipment deployed in each power grid node in the power system, wherein the measurement value data comprises voltage data, current data, temperature monitoring data and tide monitoring data;

passing the measured value data of each power grid node through a context-based encoder model comprising an embedded layer to obtain a plurality of feature vectors, and cascading the plurality of feature vectors to obtain a measured value feature vector of each power grid node;

acquiring all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all power grid nodes in the power system;

respectively passing all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all power grid nodes through the context-based encoder model comprising the embedded layer to obtain a plurality of voltage eigenvectors, a plurality of current eigenvectors, a plurality of temperature eigenvectors and a plurality of power flow eigenvectors;

respectively calculating the distance between every two eigenvectors in the plurality of voltage eigenvectors, the plurality of current eigenvectors, the plurality of temperature eigenvectors and the plurality of load flow eigenvectors to obtain a voltage distance eigenvector matrix, a current distance eigenvector matrix, a temperature distance eigenvector matrix and a load flow distance eigenvector matrix;

calculating a position-weighted sum of the voltage distance characteristic matrix, the current distance characteristic matrix, the temperature distance characteristic matrix and the power flow distance characteristic matrix to obtain a parameter topological characteristic matrix;

two-dimensionally splicing the measured value eigenvectors of each power grid node into a measured value characteristic matrix;

passing the measurement value matrix and the parameter topological matrix through a graph neural network, wherein the graph neural network generates a measurement value topological characteristic matrix containing measurement value characteristic information and parameter topological characteristic information through learnable neural network parameters, and each row vector in the measurement value topological characteristic matrix corresponds to a measurement value topological characteristic vector of one power grid node;

calculating a label category evolution factor between the measured value topological feature vector of each power grid node relative to the measured value feature vector of the power grid node as a loss function value corresponding to each power grid node, wherein the label category evolution factor is a difference value between a cross entropy value between the measured value feature vector and the measured value topological feature vector and a logarithmic function value of a compressed category probability value of the measured value feature vector divided by a category probability value of the measured value topological feature vector, and the compressed category probability values of the measured value feature vector and the measured value topological feature vector are obtained by a classifier; and

training the encoder model and the graph neural network with a weighted sum of the loss function values for each of the grid nodes as a global loss function value; and

an inference phase comprising:

passing the measured value data of each power grid node through the context-based encoder model containing the embedded layer and trained by a training module to obtain a plurality of feature vectors, and cascading the plurality of feature vectors to obtain a measured value feature vector of each power grid node;

respectively passing all voltage data, all current data, all temperature monitoring data and all power flow monitoring data of all power grid nodes through the context-based encoder model containing the embedded layer trained by a training module to obtain a plurality of voltage feature vectors, a plurality of current feature vectors, a plurality of temperature feature vectors and a plurality of power flow feature vectors;

passing the measurement value matrix and the parameter topological matrix through a graph neural network trained by a training module, wherein the graph neural network generates a measurement value topological characteristic matrix containing measurement value characteristic information and parameter topological characteristic information through learnable neural network parameters, and each row vector in the measurement value topological characteristic matrix corresponds to a measurement value topological characteristic vector of one power grid node;

and respectively enabling the measured value topological characteristic vector of each power grid node to pass through a classifier to obtain a classification result, wherein the classification result is used for representing the compression ratio of each power grid node.

7. The monitoring method of a monitoring system of a power system secondary device according to claim 6, wherein passing measurement data items of each of the grid nodes through a context-based encoder model including an embedded layer to obtain a plurality of feature vectors, and concatenating the plurality of feature vectors to obtain a measurement feature vector of each of the grid nodes, comprises:

respectively enabling the measured value data of each power grid node to pass through an embedded layer of the encoder model so as to convert each measured value into an input vector to obtain a sequence of input vectors;

passing the sequence of input vectors through a converter of the encoder model to obtain the plurality of feature vectors; and

cascading the plurality of eigenvectors to obtain a measured value eigenvector of each of the grid nodes.

8. The monitoring method of the monitoring system of the power system secondary device according to claim 6, wherein calculating distances between each two eigenvectors of the plurality of voltage eigenvectors, the plurality of current eigenvectors, the plurality of temperature eigenvectors, and the plurality of power flow eigenvectors to obtain a voltage distance eigenvector matrix, a current distance eigenvector matrix, a temperature distance eigenvector matrix, and a power flow distance eigenvector matrix, respectively, comprises:

9. The monitoring method of the monitoring system of the power system secondary device according to claim 6, wherein calculating a label class evolution factor between the measured value topological feature vector of each of the grid nodes with respect to the measured value feature vector of the grid node as the loss function value corresponding to each of the grid nodes includes:

calculating a label category evolution factor between the measured value topological characteristic vector of each power grid node relative to the measured value characteristic vector of the power grid node as a loss function value corresponding to each power grid node according to the following formula;

the formula is:

10. The monitoring method of the monitoring system of the power system secondary device according to claim 6, wherein calculating a label class evolution factor between the measured value topological feature vector of each of the grid nodes with respect to the measured value feature vector of the grid node as the loss function value corresponding to each of the grid nodes includes:

respectively passing the measured value topological feature vector and the measured value feature vector through a classifier, wherein the classifier respectively processes the measured value topological feature vector and the measured value feature vector to generate a compressed class probability value of the measured value topological feature vector and a compressed class probability value of the measured value feature vector according to the following formulas:

softmax{(W_n,B_n):…:(W₁,B₁)|X}。